discussion AMD Phenom X4 9750 Quad Core CPU Review

[AkG]

AMD N3 / MetaBench

AMD n3

This benchmark was created by AMD to show what their then ultra efficient design CPU’s could really do against the older Intel P4’s. While things have changed in the computer industry this benchmark still stresses efficiency over brute force and it will be interesting to see how the new Phenom K10 architecture stands up against the Intel Core 2 Duo lines.

​

Lets face it, the n3 benchmark is ancient, but it still has to be a little bit embarrassing for AMD to be beaten at a benchmark solely built and designed to highlight how efficient their processors were in comparison to Intel CPUs! Not only loosing but being smacked around by a (relatively) ancient C2D dual core really adds insult to injury and does underscore how much tweaking the K10 architecture really needs if it is going to be able to go toe to toe with the reigning champ.

MetaBench

Another golden oldie is MetaBench. This benchmark consists of 55 different tests designed to measure performance of the whole system as unit rather than an individual part. These benchmarks consist of both “synthetic” and “practical” and they include Raytrace Rendering, Ogg Vorbis audio encoding/decoding, Blowfish encryption/decryption as well as many others. All options were left at their default levels.

​

The reason I like Metabench so much is that it can’t multitask for spit and thus the efficiency of a given core (regardless of the number of them) can easily be shown. A processor cannot hide or obscure or obfuscate the issue by using multiple cores to get the job done. When the Phenom goes head to head with even the older C2D dual or even the k8 architecture which it replaced it gets its proverbial head handed to it.

 

[AkG]

STARS Euler3D / ScienceMark 2

STARS Euler3D

This benchmark is a computational fluid dynamics (CFD) benchmark that can be tweaked to use either a single core or X number of cores. Computational fluid dynamics is extremely floating-point intensive and do a great job at showing how powerful a given architecture really is. In this instance we are using the AGARD 445.6 aeroelastic test wing available at Computational AeroServoElasticity (CASE) Laboratory in the Department of Mechanical and Aerospace Engineering at Oklahoma State University’s website (the creators of the this benchmark). This benchmark uses the “...standard aeroelastic test case used for validation of unsteady, compressible CFD codes. The CFD grid contains 1.23 million tetrahedral elements and 223 thousand nodes . . . . The benchmark executable advances the Mach 0.50 AGARD flow solution. A benchmark score is reported as a CFD cycle frequency in Hertz.”

CFD really is a great way to push a CPU to its limits. The more horsepower you throw at it, the more it wants! Over the years CFD times have dropped from days to hours and now even to minutes; it really is a wonderful time to be a computer enthusiast. Unfortunately, from these numbers it is not such a great time to be a AMD zealot as the Phenom’s efficiency really is lacking.

ScienceMark 2

The ScienceMark benchmark program attempts to measure real-world performance by focusing specifically on CPU and memory performance without involving the graphics card. This laudable goal is accomplished by using multiple tests, each of which are designed to simulate the a unique type of calculations and memory accesses used in high-performance scientific and engineering computing. These benchmarks include a cryptographic algorithm benchmark, a molecular dynamics test and even a quantum mechanics equations test. All in all it gives the CPU and memory controller quite the work out and unlike most other synthetic benchmarks its results can in no way be swayed by the power (or lack thereof) of your graphics card(s).

​

Once again AMD is straight at the end of the pack here and it is really unfortunate to see this. That being said, this test does not seem to take buch advantage of all four cores.

 

[AkG]

3DMark Vantage / 3DMark 2006

3DMark Vantage

3DMark Vantage is the latest release by Futuremark, creators of the 3DMark suite. This program is the first Futuremark version of 3DMark designed exclusively for Windows Vista. 3DMark Vantage consists of 2 CPU and 2 GPU tests as well as and 6 feature tests all of which are very hardware intensive. Four presets are available to allow for those with older PCs to benchmark just as easily as those with cutting edge hardware. For our testing, we will use all settings set at their default values.

Believe it or not, losing to the Q6600 here is not a bad thing. Its not a good thing either, but the important thing to understand is that the difference is less than 8%. With some heavy tweaking AMD can easily turn this around and make the 9x50s true winners.

3DMark 2006

For all you Vista abstainers here is the last of the 3dMark XP line. Futuremark 3DMark 06 has been the worldwide standard in advanced 3D game performance benchmarking for a few years now. A fundamental tool for PC users and gamers, 3DMark06 uses advanced real-time 3D game workloads to measure PC performance using a suite of DirectX 9 3D graphics tests, CPU tests, and 3D feature tests. For our purposes we will be focusing in on the CPU tests. All settings were once again left at their default value.

​

Unfortunately things are not as rosy in XP land as they were in Vista. For 3DMark 2006 the lower 3000 level is where high end dual cores reside and not where Quads belong. Once again AMD needs to spend a lot of money/time/effort and tweak the K10 ASAP or they are going to lose their core base of customers for another upgrade cycle and I honestly don’t know how many more bad cycles AMD can handle.

 

[AkG]

SiSoft Sandra / PCMark Vantage

Sandra Memory Latency, Processor Multi-Media, and Processor Arithmetic

The software version used for these tests is SiSoftware Sandra Professional Home XII.SP2 and the three benchmarks used are the Memory Latency, Processor Multi-Media, and Processor Arithmetic benchmarks. These three benchmarks were chosen as they provide a good indication of three varying types of system performance. The latency test shows performance of the memory controller, the multi-media test shows how the processor handles multi-media instructions and data, and the arithmetic test shows how the processor floating point instructions. These three tests illustrate three important areas of a computer’s speed and provide a wide scope of results. For our purposes we are hoping they can shed so light on the lack luster performance of the Phenom 9750 so far.

​

As we suspected, the SuperPi and other memory intensive benchmarks are being hindered by the Phenom’s less than stellar memory latency. A good chunk of this problem can be laid at the feet of the relatively slow 2MB of L3 cache. The rest can be blamed on the added latency incurred by the hardware TLB errata workaround and the fact that the memory controller is brand new which means it is in need of serious tweaking.

However, on the positive side the memory latency is not as terrible as the TLB error prone B2s are and the Phenoms MFLOPS scores are very nice. This tells us that while the core architecture needs tweaking it does have the potential to be a great architecture in the future and maybe even better than the venerable k8 was!

PCMark Vantage

The latest iteration of the popular system benchmark is PCMark Vantage from the Futuremark crew. The PCMark series has always been a great way to either test specific areas of a system or to get a general over view of how your system is performing. For our results, we simply run the basic benchmark suite which involves a wide range of tests on all of the sub-systems of the computer.

​

A difference of 27 points is nothing and this is very good news as PCMark Vantage is supposed to be a benchmark for showing what Joe and Jane SixPack will see when they use their computer in day to day life. For all intents and purposes according to this benchmark, most people will not see any difference in performance between a Q6600 and a Phenom 9750.

 

[AkG]

Game Performance: World in Conflict

Game Performance: World in Conflict

The World in Conflict in-game benchmark is a great test to show video card performance in real gaming situations. Under the Graphics menu in options, you can choose a variety of video settings and there is a "Run Benchmark" button. The actual benchmark uses all of the game’s graphic capabilities and is a good indication which settings will be optimal for a user’s system. For the tests below resolution was set to 1024x768 and graphics were left at their default values with 8XCSAA for one run and then disabled for another.

​

Buying a quad core for gaming is still counter-intuative as even the best quads lag far behind their high speed dual core breathren. Making things even worse (for the scope of this review) a Phenom 9750 does not even qualify as one of the best quad cores when it comes to gaming. In WiC’s benchmarking program it did not matter one iota whether AA was enable of not as the Phenom was consistent in posting the worst scores of any CPU tested.

 

[AkG]

Game Performance: Company of Heroes

Game Performance: Company of Heroes

The Company of Heroes is anther great game which comes with an in-game benchmark. Just like WiC, CoH is a great test to show video card performance in real gaming situations. Under the Graphics menu in options, you can choose a variety of video settings and there is a "Run Benchmark" button. The actual benchmark uses all of the game’s graphic capabilities and is a good indication which settings will be optimal for a user’s system. For the tests below resolution was set to 1024X768 and graphics were left at their default values.

​

While the Phenoms CoH scores are not as bad as the WiC game scores they are certainly nothing to get excited about either. On the bright side at least with 8XCSAA enabled the Phenom was able to beat its older (and now cheaper) but “faster” dual core k8 brother.

 

[AkG]

Efficiency / Thermal Output

Efficiency

To us efficiency is the amount of power a given processor requires to do its job. Today’s processors are certainly becoming greener as they sip power levels so low they would have been nothing more than flights of fancy just a few years ago. It will be interesting to see how the unrefined and brand new k10 architecture stacks up when compared to both AMD highly refined K8s and Intel’s C2Ds.

​

Load levels and idle levels for the Phenom while not great but are not that terrible either. Overall, what these numbers tell us is that AMDs Phenoms are not refined and that AMD needs to take the necessary steps to make their quad core processors as environmental friendly as possible or they will be relegated to a niche market. However, if rumors of the die shrunk Phenoms are correct the k10 architecture may end up being the greenest of them all; so things maybe changing for the greener...soon.

Thermal Output

We could post a fancy graph showing exactly how hot a this Phenom “fireball” 9750 really is, but it would not exactly be fair; after all, it is a 125W TDP CPU and comparing it to cooler running chips is not an "apples to apples" comparison. For anyone who has not read the R-E-S-P-E-C-T section, you can take it on faith that this is not a cool runing chip and really shouldn’t be considered appropriate for HTPCs or any other less than optimal cooling situations. This core can in fact be considered a great candidate for both water cooling or even stress testing a CoolIT Boreas (and yes that is foreshadowing and you can expect a Boreas review in the future)!

 

[AkG]

Overclocking

Overclocking

Overclocking locked Phenoms is not as easy as overclocking so called Black Edition Phenoms as you can only adjust the HT speed, HT Mulplier and various voltages. In either case, while not exactly difficult it is more involved as the built in memory controller can in some cases be the limiting factor (requiring voltage increases of its own). After working with this 9750 we were able to get a 2hr Prime 95 stable overclock of 2.83Ghz.

To achieve this we had to lower HT multiplier quite a bit but we have a hunch this issue is more motherboard related than CPU as the DFI wanted to auto default the HT to 5x even at stock speed. All in all we were able to get a 430MHz speed increase or just about 18% overclock. For such an early architecture this is certainly not a terrible achievement.

Over the years we as an overclocking community have become spoiled and expect to get massive overclocks from our processors; this increased level of expectation is partially AMDs fault with their mature k8 architecture. However, it is mainly because of the highly refined C2D line of Intel processors which were amazing right from the get go for their jaw dropping overclocking ability. As stated, 2.83GHz is not bad and we were satisfied with this result.

​

To see how much of an improvement this would have over stock scores we reran some of our benchmarks to get a feel for what the future may have in store for the K10 Phenom line. This is what we found out.

​

As you can see Overclocking the Phenom results in a mixed bag of performance boosts. Some benchmarks return only 17-19% increase for our 18% overclock while others net a nearly doubling of scores.

To us this really underscores how uneven and unrefined the 9750s are. Please don’t get us wrong, they have horsepower to burn but they are still not as elegant as the K8 was. We look forward with anticipation to seeing what this processor can develop into, but for right now even overclocking nets less than ideal performance gains.

 

[AkG]

Conclusion

Conclusion

As with most things in life its better to get the bad news over and done with first, and in the case of the Phenom 9750 it is pretty bad. With all the various tests and benchmarks we threw at it, the Phenom did not perform especially well in any area. It didn't matter what we looked at; whether it be thermal load, game performance, or even efficiency the Phenoms lagged behind the Q6600, and when we included the Q9450 things get down right grim. Sure there were a few highlights along the way but they were more glimpses of the future and only underscored how much work and revision the K10 architecture needs if it is to compete on an even footing with Intel.

The good news is that the K10 Quads have pushed AMD back into the game and we can now actually consider them seriously again in some cases. The Phenom line really is a wonderful, if unrefined piece of kit. Unfortunately, a lot of people get caught up in the marketing mumbo jumbo and forget that even a quad that can’t beat Intel’s is more than Joe and Jane SixPack will ever need (or at least need for the next two years). After working and using this processor for a few weeks now, its numbers really do belie how good it is at day to day tasks. That being said needing a high end cooler does ruin the effect somewhat, but the 9750 is still a wonderful processor that you should take a second look at if you are looking for a mid-range AMD system.

This leads us to what we think AMD needs to do. AMD needs to regroup and do exactly like their ATI division is doing: concentrate on the mid end and forget about the high end. Now we are not saying give up on the high end but rather not worry about being top dog for right now. Being the top dog may be great for company morale and we're sure it warms the cockles off their fans' hearts, but it doesn’t do much for the bottom line. The middle of the road is where the big bucks are, as most consumers buy mid-grade computers and are happy as clams with them. Basically, they need to cut costs which will in turn yield more savings for the consumers. If they are able to deliver a competitive quad core which costs less than the Q6600, they will be half way there. Unfortunately, at this time the Phenom just feels like a rushed product that is suffering from some flaws which should have been caught early in tha game.

Besides the above, the biggest thing they have to do is get on the manufacturer's cases and push them to put out decent motherboards which can handle the high end Phenoms in the $100 - $150 range. As it stands, choosing a motherboard that supports a Phenom is a bit of a bewildering process which only serves to turn many customers to the Intel camp.

This to us is the short term stop loss maneuvers that need to happen. The next thing they have to do is get the efficiency and thermal loads under control, and get some all round refinement to the K10 happening ASAP. If AMD can do this then they can recapture a good portion of the HTPC, Silent PC and gaming PC crowd. These are the consumers who want cool-running CPUs which leave lots of juice for all the bangs and whistles (like quad-fire / quad SLI, or loads of hard drives) yet can still crunch massive amounts of data when need be. This will also make air cooling w/ reasonable core temps more possible as no one really wants to watercool a chip in the mid range market (unless that is your hobby).

All in all, it seems like AMD does have a fighting chance in the quad core arena but they will have to move fast to make up for lost ground. The Phenom X4 9750 seems to be a good second step after the lackluster B2 chips but it is far from a solid performer and is still seriously lacking in several other key areas. Rumors speak of new, cooler running Phenom’s coming out soon so maybe, just maybe we are unto something and we can look forward to a real two horse race again in 2009. But until then, the fact is that AMD seems to be two steps or more behind Intel while not offering anything new in the way of value.


Pros:
- Quad core AMD!
- More than enough power for the average computer user
- Reasonable price
- Shows that the K10 architecture has true potential


Cons:
- Subpar performance when compared to Intel Quads
- HOT running
- Requires a good board (or else bad things happen!)
- Not as efficient as the Intel Quads
- K10 needs refinement
- Latency Issues